Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
WHO/HSE/GAR/BDP/2009.2 Rift Valley fever outbreaks forecasting models Joint FAO - WHO experts consultation Rome, Italy 29 September–1 October 2008 GLOBAL ALERT AND RESPONSE Rift Valley fever outbreaks forecasting models Joint FAO - WHO experts consultation Rome, Italy 29 September–1 October 2008 Food and Agriculture Organization of the United Nations GAR publications are available oline at: www.who.int/csr/resources/publications © World Health Organization 2009 All rights reserved. The designations employed and the presentation of the material in this publication do not imply the expression of any opinion whatsoever on the part of the World Health Organization concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. Dotted lines on maps represent approximate border lines for which there may not yet be full agreement. The mention of specific companies or of certain manufacturers’ products does not imply that they are endorsed or recommended by the World Health Organization in preference to others of a similar nature that are not mentioned. Errors and omissions excepted, the names of proprietary products are distinguished by initial capital letters. All reasonable precautions have been taken by the World Health Organization to verify the information contained in this publication. However, the published material is being distributed without warranty of any kind, either express or implied. The responsibility for the interpretation and use of the material lies with the reader. In no event shall the World Health Organization be liable for damages arising from its use. This publication contains the collective views of an international group of experts and does not necessarily represent the decisions or the policies of the World Health Organization. Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 Contents Executive Summary ....................................................................................................2 I. Introduction ..............................................................................................................3 II. What is needed to improve RVF forecasting and outbreak management? ........4 Return from experience from recent outbreaks and expectations from FAO and WHO ...................... 4 III. The Natural History of RVF ....................................................................................5 Epidemiology and ecology of RVF and key drivers for endemicity and epidemics .............................. 5 Return from experience: Dambo versus irrigation schemes ................................................................ 6 Data, tools and methods for the identification of RVF risk areas: experience from the Horn of Africa 8 The example of eastern Africa: the dynamic of RVF and tools for monitoring virus activity ................ 8 Studying the relationship between environmental factors retrieved from remotely sensed images and RVF in Senegal, western Africa............................................................................................................ 9 IV. RVF risk distribution maps and forecasting models: what is currently available and how can it be improved?.................................................................9 Generating RVF suitability maps and predicting RVF distribution: application and limits .................... 9 Monitoring NDVI anomalies and predicting potential RVF activity ..................................................... 10 V. What other techniques could be applied to improve RVF forecasting? ..........11 Real-time identification of flooded areas ............................................................................................ 11 The Indian Ocean dipole in 2006–2008.............................................................................................. 12 The role of wind in the spread of RVF ................................................................................................ 12 VI. Discussion ............................................................................................................14 VII. Recommendations ..............................................................................................17 Acronyms ............................................................................................................................................ 20 Participants, observers and FAO-WHO Secretariat................................................21 –1– Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 Executive Summary Rift Valley fever (RVF) is a mosquito-borne viral zoonosis that primarily affects animals but also has the capacity to infect humans. RVF outbreaks in eastern Africa are closely associated with periods of heavy rainfall and RVF forecasting models and early warning systems have been developed to enable national authorities to implement measures to avert impending outbreaks. Despite these precautions a series of RVF outbreaks from 2006 to 2008 dramatically affected countries in Africa, the Sudan and islands in the Indian Ocean. Based on practical considerations experienced during 2006–2008 RVF outbreak control operations, the Food and Agricultural Organization of the United Nations (FAO) and the World Health Organization (WHO) have developed a common strategy for RVF outbreaks, from forecasting to response. FAO and WHO are currently defining joint guidelines for countries, to provide a framework for a coordinated and integrated prevention and control strategy, before, during and after RVF outbreaks. Part of this effort must be dedicated to early warning based on RVF outbreak modelling and forecasting, which is critical for providing a time window for preventive measures before the amplification of the virus is out of control. To further this effort, FAO and WHO invited a group of experts on RVF modelling and forecasting to a two-day Rift Valley fever outbreak forecasting models workshop to share feedback from the 2006– 2008 outbreaks, share experiences, identify gaps and explore potential improvements in RVF outbreak models. The objectives of the workshop were to review the natural history of RVF, review the forecasting models and risk distribution maps available and being developed, and propose how these tools might be improved. The final goal was to define a roadmap, in the form of recommendations, for the development of tools for forecasting RVF outbreaks and for real-time analysis of RVF spread during outbreaks. The following report presents a summary of the presentations of different RVF forecasting models and RVF outbreak risk mapping strategies with possible improvements identified by the speakers, the main points of discussion and the concluding recommendations based on the points listed below. 1. The accuracy of RVF potential major outbreak area maps should be increased in order to improve forecasting models. 2. Forecasting models should be further developed to be more specific. 3. Models should be coupled to buy time and precision; an alert signal should be sent six months before the start of an animal outbreak. 4. RVF forecasting models and potential major outbreak area maps should be used to monitor evolution of RVF outbreaks over time. 5. The participation of ministries of meteorology, ministries of health and ministries of agriculture in the forecasting alert system should be enhanced. –2– Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 I. Introduction Rift Valley fever (RVF) is a viral zoonosis that was first identified in Kenya in 1931. This mosquitoborne disease primarily affects animals but also has the capacity to infect humans. The majority of animal infections result from the bites of infected mosquitoes, while most human infections are caused by direct or indirect contact with the blood or organs of infected animals. Such contact may occur during the care or slaughtering of infected animals or possibly from the ingestion of raw milk. RVF outbreaks in eastern Africa are closely associated with periods of heavy rainfall that occur during the warm phase of the El Niño/Southern Oscillation (ENSO) phenomenon, which affects the mosquito populations acting as vectors and reservoirs of the disease. Forecasting models and early warning systems for RVF using satellite images and weather/climate forecasting data have been successfully developed, enabling national authorities to implement measures to avert impending epidemics. In late 2006 and the first semester of 2007, RVF outbreaks linked to unusual rainfall associated with El Niño, dramatically affected countries in the Horn of Africa (Kenya, Somalia, and the United Republic of Tanzania). The disease reappeared in the Sudan in September 2007 following excessive rainfall driven by a post-El Niño, unusually warm, sea temperature in the Indian Ocean. Southern African countries (Swaziland and South Africa) and islands in the Indian Ocean (the Comoros, Mayotte, and Madagascar) were affected in 2007 and 2008. The outbreaks were predicted for the coastal countries of eastern Africa. Forecasting models and early warning systems based on near-real-time climatic data were available at the continental level and proved to be efficient in raising the alert before the onset of the epidemic. Warning messages were sent, but the implementation of preventive measures and key control activities in this context was difficult for various reasons. FAO and WHO collaborated in the emergency response effort with other international organizations including the Centers for Disease Control and Prevention (CDC, Atlanta, USA), the Institut Pasteur, ( Madagascar, and Paris, France), the World Organisation for Animal Health (OIE, Paris, France) and the United Nations Children’s Fund (UNICEF); national organizations; and the national authorities of the affected countries. National action plans were reviewed, disease control activities implemented and a significant amount of resources allocated to increase disease detection surveillance and control. As a result, unique human and animal data were collected which can now be used to adjust and complement the existing RVF forecasting models. In addition, FAO and WHO officers in the field had the opportunity to widely use outcomes from the different models and identify gaps or needs that could be filled in order to improve the use of these predictions in an integrated disease prevention and control perspective. FAO and WHO are currently working on the definition of joint guidelines for countries to improve the level of preparedness and capacity for early response to RVF outbreaks. Part of this effort must be dedicated to early warning based on RVF outbreak modelling and forecasting as it is a key point that may provide a time window for preventive measures, before the amplification of the virus is out of control. To further this effort, FAO and WHO held a Rift Valley fever outbreak forecasting models workshop, 29 September–1 October 2008. RVF modelling and forecasting experts were invited to share feedback from the 2006–2008 outbreaks, share experiences, and identify gaps and explore potential improvements in RVF outbreak models in order to further adapt model outcomes to fit the needs of RVF control strategies. The objectives of the workshop were to review the natural history of RVF, review the forecasting models and risk distribution maps available and being developed, and propose how these tools might be improved. The final goal was to define a roadmap, in the form of –3– Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 recommendations, for the development of tools for forecasting RVF outbreaks and for real-time analysis of RVF spread during outbreaks. II. What is needed to improve RVF forecasting and outbreak management? Return from experience from recent outbreaks and expectations from FAO and WHO Dr Stéphane de La Rocque, FAO, Rome and Dr Pierre Formenty, WHO, Geneva The first warning of an increase in the probability of RVF activity at the end of 2006 was due to the development of an El Niño event and was launched by the NASA/Goddard Space Flight Center in mid-September of the same year. FAO and WHO forwarded this warning to their representatives in the at-risk countries and consolidated warnings were sent in November. However, too short notice, limited confidence of the countries in this sort of prediction, a decrease in awareness and resources dedicated to RVF and the absence of clear regional strategies and common contingency plans at the country and regional levels made the implementation of preventive measures difficult. Based on practical considerations experienced during RVF outbreak control operations conducted in some of the affected countries - mainly Kenya, Madagascar, Somalia, the Sudan, and the United Republic of Tanzania - FAO and WHO have developed a common strategy 1 for RVF outbreaks, from forecasting to response. FAO and WHO are currently defining joint guidelines for countries to provide a framework for a coordinated and integrated prevention and control strategy before, during and after RVF outbreaks. FAO and WHO have the common goal of supporting development of the RVF forecasting effort to improve the capacities of existing models in order to increase the period in which preventive measures can be undertaken. For various reasons, RVF forecasting is more advanced in some countries than others; RVF high-risk ecological areas are well understood in Kenya, Somalia, and the northern part of the United Republic of Tanzania, and less well identified in the Sudan. Areas where the disease had never been reported, such as the centre of the United Republic of Tanzania, were affected during the recent epidemics. Forecasting is still exploratory in some countries, such as Madagascar, due to a limited understanding of the epidemiology of RVF in the different ecosystems of the island. The role of animal movements in the spread of the disease makes it necessary to distinguish between the primary area of emergence and secondary foci where the virus has been introduced and amplified, especially in irrigated areas, e.g., the Gazeera irrigation scheme in the Sudan, the Ifakarra rice valley in Tanzania or rice valleys in the centre of Madagascar. There is therefore a real need to further improve the spatial accuracy of forecasts and monitor conditions in regions of interest in order to deploy resources and strengthen efforts for active surveillance detection and control. FAO and WHO have identified, in practical terms, some of the limitations of and remaining questions for forecasting activity: 1 P Formenty and S de La Rocque. Rift Valley fever outbreaks in Africa and the Middle East: forecasting, surveillance and control. In: Rift Valley Fever Workshop, January 27–29, 2009, Cairo, Egypt (http://www.ars.usda.gov/meetings/rvf2009/presentations.htm) –4– Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 x When should FAO and WHO alert the countries? How should the NASA RVF monthly maps of areas at risk for RVF outbreaks be interpreted when only one big outbreak (2006–2007) has been observed in seven years and when strong warnings can be highly damaging? x Could the modelling framework be improved with good field data such as soil type, elevation, vector ecology maps, etc.? x How can we address the fact that climatic models apply well in eastern and southern Africa, but are more difficult to apply to western Africa (Senegal, Mauritania), the Middle East (Egypt, Saudi Arabia, Yemen) or Madagascar? Will this affect the confidence we have in the existing models? x RVF warnings are given two months in advance, but six months are needed between the forecasting alert and outbreak onset in order to implement preventive measures, including social awareness and mass animal vaccination. Is this feasible? III. The Natural History of RVF Epidemiology and ecology of RVF and key drivers for endemicity and epidemics Dr Glyn Davies RVF endemic cycles occur in temperate, tropical and sub-tropical zones of Africa. The virus is capable of inhabiting a variety of different bioclimatic conditions including wet and tropical areas such as Côte d’Ivoire and the Congo, hot and arid areas such as Yemen or Chad, and irrigated regions such as the Senegal River valley and the Nile Delta. Most RVF viral activity is cryptic at a low level and not associated with detectable disease in humans and animals. Many African countries have found significant seroprevalence in sheep, goats and cattle for the RVF virus throughout various agroclimatic zones, without clinical signs being reported in humans or animals. Most countries are not really aware of the circulation of the virus because of a lack of systematic surveillance activities. At irregular intervals of about 5–12 years, large epidemics of RVF have occurred in southern and eastern Africa; these epidemics have been associated with above average rainfall, a large amount of vector activity and the presence of susceptible livestock. The 1997–1998 epidemic in the Horn of Africa, which is considered as one of the most devastating RVF epidemics in eastern Africa, was associated with torrential rains - 60–100 times the seasonal average - that resulted in the worst flooding in the Horn of Africa since 1961. Flood plain zones are particularly affected, as are any areas where the water table has been raised by irrigation or other water conservation practices, and serve as breeding habitats for RVF mosquito vectors. Because climatic events tend to occur over large areas, there is a tendency for outbreaks to occur simultaneously in adjacent territories. Throughout its wide ecological range, the pattern of RVF activity is much influenced by the ecological characteristics of the particular biotope under consideration. African agro-climatic zones can be used as a good guide of RVF virus activity, summarized as follows: (a) enzootic activities (+/-cryptic) in high rainfall forests and forest edges; (b) periodic increased virus activity to epizootic proportions in bushed and wooded grasslands; (c) rare but explosive epizootic RVF activity in dry grassland and semi-arid zones associated with flood plains. –5– Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 Return from experience: Dambo versus irrigation schemes Dr Pierre Formenty, WHO, Geneva and Dr Stéphane de La Rocque, FAO, Rome Spread of RVF from endemic areas can occur with the movement of livestock and the introduction of viraemic animals in conducive areas, such as irrigation schemes. At the primary foci site the virus is maintained in the vectors and/or hosts. At the secondary foci the virus is imported and subsequently spread between naïve ruminants, thanks to local mosquitoes that are competent vectors for RVF such as certain Culex and Anopheles species. During the 2007–08 outbreak in the Sudan, the disease reached the irrigated area of Gazeera through sheep, camels and livestock on their way to the capital and the northern coast, from where several hundred thousand sheep are exported to Mecca for the Hajj. Consequently, a new agro-climatic zone should be added to those described above by Glyn Davies: irrigation scheme areas where very rare but long lasting RVF outbreaks in animals and humans occur following introduction of viraemic animals. High densities of vectors and the abundance of naïve ruminant populations, including highly susceptible exotic dairy cattle, acted as facilitating factors for the amplification of the virus, with a high impact on humans. The same applied for the Ifakarra rice valley in Tanzania, 2007, and in Madagascar in 2007–08. In Madagascar the cases were initially detected in the highlands, but the infection may have been imported from the livestock breeding areas in the south and north-west when the animals were transported to meat markets around the capital or to be used as oxen in the rice fields, where the mosquitoes are abundant. These different epidemiological processes are schematically illustrated in Fig. 1. –6– Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 Fig 1. Schematic transmission of Rift Valley fever (P. Formenty, WHO): Box A, primary foci: 1) Following heavy rainfall, dambos with Aedes breeding sites are flooded for a period of at least 10– 15 days. 2) A portion of infected Aedes eggs cease dormancy and hatching adults infected with RVF virus emerge. 3) Female Aedes mosquitoes are aggressive mammophilic blood-feeders; they bite surrounding domestic ruminants such as cattle, camels, sheep and goats that serve as amplification hosts for the virus. An RVF animal outbreak is characterized by massive abortion in pregnant females and death of young ruminants 4). The large majority of human infections result from direct or indirect contact with the blood or organs of infected animals during slaughtering or butchering, assisting with animal births, delivering veterinary care or handling fresh carcasses or fetuses. 5) Latter in the course of the outbreak, secondary vectors such as Culex, breeding in stagnated waters, acquire the virus during a blood meal on an infected animal and then contribute to the transmission of RVF virus to humans. The + within the circles indicates RVF infected sheep. Box B, secondary foci: 6) Domestic animals infected with RVF at the primary site are transported to irrigation areas. 7) Irrigation schemes are persistent habitats for Culex, Mansonia and Anopheles mosquitoes, whose abundant populations increase the circulation of the virus between amplifying hosts (cattle, camels, sheep and goats). 8) As in primary foci, human infections may result from direct or indirect contact with the blood or organs of infected animals. 9) However, most of the human infections result from the bites of an infected mosquito that acquired the virus during a previous blood meal on an infected animal. –7– Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 Data, tools and methods for the identification of RVF risk areas: experience from the Horn of Africa Dr Hussein Gadain, USGS FEWS NET, Currently with FAO SWALIM, and Dr Glyn Davies Models for determining areas at risk for RVF outbreaks must consider several factors. Poorly drained soils are prone to flooding and provide good habitats for floodwater Aedes mosquito species. Surface water hydrology is influenced by rainfall, soil type, vegetation type, humidity, and temperature. A RVF risk model for the detection of areas at risk for an outbreak of RVF was developed by Hussein Gadain, Paul Ruwambo, Maxx Dilley and Glyn Davies for an International Research Institute for Climate and Society (IRI) project funded by the United States Agency for International Development (USAID) in collaboration with the United States Geological Survey Famine Early Warning Network (USGS FEWS NET), the African Union/Inter-African Bureau for Animal Resources (AU-IBAR), the Red Sea Livestock Trade Commission (RSLTC) and others. With the model, geographical characteristics of areas where RVF virus activity has been recorded in the past were analysed and correlated with the following: vegetation, considering the typology of ecotypes, from low to high potential for RVF epidemic; soil types, from poorly drained to fluvisols and gleysols; and topography, considering elevation and drainage systems. Streams and drainage channels, stream flow and flood forecasting using MM5 2 , ETA 3 and other models for the Horn of Africa were initiated in 2002 using short term climate information forecasting. Possible improvements: The next phase of the RVF risk model development should involve the temporal analysis of rainfall time series data from 1960 to the present, the determination of thresholds for soil moisture and vegetation/normalized difference vegetation index (NDVI), and the use of basin excess rainfall maps (BERM). A real-time rainfall follow-up system using moderate resolution imaging spectroradiometer (MODIS) data could subsequently be integrated into the model in order to highlight flooded areas. The potential of the existing model of river runoff and irrigated area dynamics in relation to rain events should be explored. Finally, it must be remembered that rain events may be distant in space from the at-risk areas, e.g., as in Yemen. The example of eastern Africa: the dynamic of RVF and tools for monitoring virus activity Dr K. Linthicum, USDA-Center for Medical, Agricultural & Veterinary Entomology RVF virus is transmitted by a very large number of arthropods. There are 38 species of mosquito that have been found infected in nature, at least 35 of which have been proven as competent vectors under controlled conditions. Six genera of mosquitoes are represented: Mansonia, Anopheles, Coquillettidia, Eretmapodites, Culex, and Aedes (including Ochlerotatus). Certain species of the latter two are considered to be the main vectors. In addition, mechanical transmission has been demonstrated with other haematophagous insects, including stomoxes, phlebotomies or Culicoides midges. The wide diversity of arthropods from which the virus has been isolated emphasizes the difficulty of understanding the epidemic/endemic cycle of the virus. Current evidence suggests that the RVF virus in sub-Saharan Africa is primarily maintained in interepidemic periods by transovarial transmission in Aedes mosquitoes. Aedes eggs can resist desiccation 2 MM5 is a weather forecasting model developed at Pennsylvania State University, USA 3 ETA is an atmospheric model available at the National Centers for Environmental Prediction (NCEP), USA. –8– Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 for long periods to hatch in water following prolonged rainfall or flooding. Aedes mosquitoes are thought to be the reservoir of the Rift Valley fever virus between epidemics triggered by aboveaverage rainfall, which leads to a rapid increase in vector numbers. Once infection has been amplified in livestock, secondary epidemic vectors such as Culex or, Anopheles mosquitoes, that breed in semipermanent pools of water and get infected by biting infected vertebrates, can become involved in transmission. Some, like Culex, serve as excellent secondary vectors if immature mosquito habitats remain flooded long enough. Studying the relationship between environmental factors retrieved from remotely sensed images and RVF in Senegal, western Africa Dr Pietro Ceccato, IRI, Columbia University Temporal analysis of RVF events in sentinel herds in relation to rainfall, vegetation index, bodies of water tracked by MODIS and Senegal River flow measurements shows that RVF transmission occurred in Senegal during periods with rainfall levels both below and above normal levels in the three different ecotypes present: river valley, steppe-savannah, and tree-savannah. Considering this, it is possible that in this part of Africa the immune status of animal herds, herd movements and local conditions favouring the reproduction of mosquitoes might play a more significant role in the inter-annual variability of RVF outbreaks. Data and tools for analysing rainfall estimates are derived from both rainfall estimates (RFE) and CMORPH 4 . NDVI and rainfall forecast products are available free of charge via the IRI Map Room web page 5 . IV. RVF risk distribution maps and forecasting models: what is currently available and how can it be improved? Generating RVF suitability maps and predicting RVF distribution: application and limits Professor David J. Rogers, Oxford University The Oxford University group favours a different approach to RVF distribution mapping with underlying assumptions of normally or multivariate normally distributed predictor variables. Oxford University uses records - spatial and temporal distributions - of the occurrence of RVF during the 20th century to define their eco-climatic limits using topographic data and remotely sensed satellite data within a discriminant analytical model framework. Generating RVF suitability maps and predicting RVF distribution begins with data derived from AVHRR-NOAA. 6 Monthly maximum value composite AVHRR Channel 3, land surface temperature 4 CMORPH is a method that produces global precipitation estimates. 5 http://iridl.ldeo.columbia.edu/maproom/. 6 AVHRR is a broad-band 4- or 5-channel scanner sensing in the visible, near infra-red and thermal infra-red portions of the electromagnetic spectrum. This sensor is carried on the National Oceanic and Atmospheric Administration’s (NOAA’s) polar orbiting environmental satellites (POES). –9– Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 (LST) and NDVI data are temporal Fourier processed to extract annual, bi-annual and tri-annual seasonal signals showing the amplitudes and phases (or timing), means, maxima, minima and variances. Spatial and temporal AVHRR data and digital elevation model (DEM) data are extracted from their datasets to correspond with RVF occurrence or absence. Non-linear discriminant analysis captures the covariance characteristics of sites of disease presence and absence and these are used to define the location, within multivariate space, of any point within the risk-mapped area. On this basis, the probability with which the point belonged to the category of disease presence or absence is calculated and this probability is entered into the final map. A set of RVF occurrence/absence data is derived using a literature search. An environmental envelope expansion approach is used in order to extract a maximum amount of information, as is done with such sparse data sets. As a result, the map shows the average predicted maximum likelihood posterior probabilities of environmental suitability for the disease, i.e., the probability with which each pixel belongs to the category of disease presence. Possible improvements: Statistical models rely on data availability. The Oxford University model could be significantly improved by completing the RVF occurrence/absence dataset (today it includes only data extracted from the scientific literature, from 1968 to 2004, and some cases referring to administrative units) and by distinguishing the primary foci (the area where the disease emerged) from secondary foci (where the disease was later introduced by animal movement). The use of Terra/Aqua MODIS data could be of relevance, as well as the addition of a rainfall layer as a potential predictor variable. Methods to incorporate seasonality in the resulting static RVF suitability maps should be explored, such as a Fourier analysis on the disease time series. Monitoring NDVI anomalies and predicting potential RVF activity Dr Assaf Anyamba, NASA Goddard Space Flight Center & Goddard Earth Sciences Technology Center (GEST/UMBC) NASA has developed a monitoring and outbreak risk mapping system that utilizes a variety of satellite measurements including sea surface temperatures, outgoing long wave radiation, rainfall and landscape ecology using the NDVI. The measurements represent the total variety of climatic and ecological drivers that would lead to conditions associated with the emergence of RVF vectors resulting in episodic patterns of epizootics/epidemics over time. These data are input into an RVF prediction system to produce a dynamic map of areas at potential risk for RVF activity. This system operates in near-real time to monitor RVF risk on a monthly basis and offers the opportunity to identify ecoclimatic conditions associated with potential vector-borne disease outbreaks over large areas. It has been in operation for the last 10 years. The NASA monitoring and prediction system relies on interpretation of the leading indicators in the context of the ENSO, rainfall and analysis of NDVI anomalies to map areas with conditions that would support RVF vector emergence, production and propagation. It incorporates two major sets of inputs: (1) global observations of sea surface temperatures (SST) and outgoing long wave radiation (OLR) as a proxy for rainfall, and SST indices for NINO3.4 7 and the Western Indian Ocean (WIO) as early 7 The NINO3.4 index is one of several ENSO indicators based on sea surface temperatures. An El Niño or La Niña event is identified if the five-month running-average of the NINO3.4 index exceeds +0.4°C for El Niño or -0.4°C for La Niña for at least six consecutive months. – 10 – Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 leading condition indicators; and (2) continental scale observations of RFE and NDVI time series measurements for Africa. The algorithm is as follows: x Potential epizootic areas are mapped, based on an RVF literature survey, to identify countries where there have been episodes of RVF activity. These maps are improved through climate variable, rainfall and NDVI threshold determination to derive what is referred to as a potential epizootic area mask (PEAM). x NDVI anomalies are calculated to identify areas of anomalous positive conditions (positive anomalies > 0.025 threshold, with reference to desert calibration); persistently positive anomalies with a three month mean > 0.1 are identified. All ‘pixels’ that meet these criteria and are within the PEAM are marked as RVF risk area pixels. All the leading global indicators and continental analyses are updated and interpreted and every month a report assessing the risk of RVF activity is issued and posted on the web site. 8 Possible improvements: The NASA system could be improved with the incorporation of rainfall into the algorithm. This input of rainfall would provide a back-up check on the NDVI anomalies and allow for the improvement of RVF outbreak risk mapping through the ranking of risk based on accumulated rainfall. It is proposed that the daily RFE rainfall product from the NOAA/Climate Prediction Center be used for this purpose. Some of the RVF outbreaks along coastal Kenya in 2006–2007, in South Africa in 2008, and in central Madagascar in 2008–2009 were outside the RVF PEAM. Determination of the PEAM needs to be improved so that areas are not missed. V. What other techniques could be applied to improve RVF forecasting? Real-time identification of flooded areas Dr Xiangming XIAO, ISE, University of New Hampshire For the real-time identification of flooded areas, a MODIS-based mapping algorithm can be applied to capture flood dynamics. Such an algorithm, based on eight-day composite MODIS data, simultaneous tracking of NDVI, land surface water index (LSWI) data and enhanced vegetation index (EVI) data has been successfully used to monitor rice production and harvest in relation to the occurrence of avian influenza in Thailand. It could be further developed, with minor modifications, to reveal flooded areas in RVF endemic areas in near-real-time mode. Adjustments will be needed to use this sort of data in Africa, but this could be done within a short period of time. Another technique that could be applied here would be to use daily MODIS data, which would significantly improve the capacity to delineate flood/inundation timing and duration. With improved flooding maps (location, timing and duration), statistical spatial analysis could be developed, using historical data on RVF outbreaks, and applied in order to better identify risk areas and risk conditions for the emergence of RVF. 8 http://www.geis.fhp.osd.mil/GEIS/SurveillanceActivities/RVFWeb/indexRVF.asp – 11 – Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 The Indian Ocean dipole in 2006–2008 Dr Chun Knee Tan, United Nations University Indian Ocean dipole (IOD) events may occur in concert with or independent of an El Niño event and have been linked to prolonged heavy rainfall in eastern Africa and subsequent positive anomalies in vegetation indices. Although prediction of IOD events is still in its infancy, IOD-related increased seasonal rainfall in eastern Africa could be forecasted at the regional level months in advance. In 2006, the IOD associated with El Niño was predicted four to six months in advance. In 2007, there was also a weak positive IOD associated with La Niña. In 2008, an IOD event occurred again and was predicted four months in advance. Improvement of the understanding of this phenomenon could be of great interest for RVF risk management, however, much remains to be done. Monitoring for IOD events needs to be further developed and the reliability of the prediction in terms of accuracy, magnitude of the event and location and duration of the consecutive climatic events needs to be assessed. In addition, the dynamics of the different IODs are not all the same; discrepancies between the different prediction models and limitations of the models need to be addressed and correlations with the occurrence of RVF need to be explored. The role of wind in the spread of RVF Dr Guy Hendrickx, Avia-Gis Recently it has been shown that wind can be an important quantifiable driver of the spread of arthropod borne diseases, such as bluetongue, in temperate regions. This work was inspired by Sellers (1980) 9 who hypothesized the importance of this route of spread in the epidemiology of arboviruses in the tropics. He argued that in the area covered by the Intertropical Convergence Zone (ITCZ), a sequence of disease endemicity zones can be discerned according to the interplay of host, vector and climate and illustrated this, amongst other examples, for Japanese encephalitis. It is possible that wind may play a role in the spread of RVF as well, although this has yet to be investigated. Proposed activities: The role of wind in the spread of RVF should be analysed. This could be done in the following manner: 9 x Spatial and temporal RVF outbreak databases (with major epidemics from 1977 to 2008) could be used as a source of data and available WHO databases matched with published information on animal outbreaks (in peer reviewed papers and OIE and FAO publications) in order to explore the role that wind may play in the spread of RVF during previous outbreaks. x A spatio-temporal analysis of outbreak data could be conducted for each outbreak. The aim of this analysis would be to identify and geolocate as precisely as possible the origin of each RVF outbreak (with relevant information on the genetic signature of the RVF strain concerned, if available), the occurrence of spatial clusters and the occurrence of spatio-temporal patterns in its spread. Sellers RF. Weather, host and vector--their interplay in the spread of insect-borne animal virus diseases. Journal of hygiene, 1980, 85:65–102. – 12 – Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 x For each outbreak, a formal forward wind trajectory analysis could be performed. To this end, the Global Data Assimilation System (GADS) database could be used as input wind data. This database is processed by the US National Weather Service’s National Centers for Environmental Prediction. The model includes parameters to determine the vertical velocity movement. The trajectories could then be calculated for each starting point at different starting altitudes, i.e., 100 m, 1000 m and 5000 m. Next to the trajectory data, meteorological data should also be calculated along the trajectories for each time step, including ground temperature and air temperature. x An analysis of the computed trajectories and matching meteorological data should be conducted in light of the chronology of RVF events during each outbreak to verify if wind could play a role in the spread of the virus. – 13 – Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 VI. Discussion The main points of discussion during the meeting are listed below: x Most RVF forecasting models are based on human case data. Indeed, very limited data on animal outbreaks are available. However, the location of the human cases may not always reflect the real place were the virus emerged, as many reports refer to hospitals where the sick were treated. Furthermore, the virus amplifies in animals, and its distribution may be much wider. x It must be borne in mind that RVF is endemic in most African countries, and that the transmission processes may differ depending on the ecosystems. A distinction must be made between virus activity, outbreak and disease. In the perspective of users - ministries of health, ministries of agriculture, FAO and WHO - the objective must focus on major epidemics. x There is a clear need to differentiate primary versus secondary cases. Primary cases reflect emerging sites where the virus may survive during an inter-epidemic period and where early detection of virus activity must be focused. Secondary cases reflect the places where the virus has been introduced and can amplify, for example, after transportation through infected animal movement, including trade, as in the RVF outbreak in the United Republic of Tanzania or in the centre of Madagascar. It is important to characterize these places in terms of disease management and risk for human health. It is well understood that the ecological drivers behind these primary and secondary cases may differ. In the perspective of early warning and risk mitigation before RVF emergence, FAO and WHO recommend that the effort should focus on primary cases. Participants request FAO and WHO to help by providing accurate data to the research community and by describing primary versus secondary cases; this means consolidating and revisiting the RVF databases, with a particular effort to include expert opinions. x There is a good understanding of the role of certain species of mosquitoes in the course of an epidemic, especially in the dambo system of eastern and southern Africa. In other parts of the continent and in other ecosystems, such as irrigated areas or forest, the picture is less clear. However, vector population distribution maps are critical to potential emergence during the inter-epidemic period and epizootic events. In the users’ perspective described above, efforts must focus on predicting the activity of primary vectors like Aedes, as opposed to secondary vectors such as Culex, Anopheles, Mansonia or other blood-feeding arthropods. Environmental conditions for mosquito development should be further described and this could contribute to the definition of primary sites. Finally, the passive dispersal of mosquitoes by winds needs to be better investigated. x The role of immunity in animals in the course of an epidemic has not yet been evaluated, and may play a role in RVF outbreaks. x RVF is focal during inter-epidemic periods. However, it is multi-focal at the beginning of an epidemic and this justifies a regional approach, or a pattern-based approach, to improve the sensitivity of the different models. In the users’ perspective, efforts should focus on identifying amplification when the disease is still focal, in characteristic ecotypes; soil information and hydrological spatial data should be included in the models. x Measurements of uncertainty in predictions should also be defined with the creation of a map of confidence levels. A way to quantify accuracy and reliability should be elaborated, so that warning messages can be put into perspective by users when disseminated. – 14 – Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 x There is a clear need to train users (ministries of health, ministries of livestock) on how to read forecasts. x The general pathway of factors leading to an RVF outbreak highlighting where map-based risk assessment and existing models come into play was drawn up (Fig.2). Participants determined a roadmap for RVF forecasting (Fig.3). Logically, the forecasting capacities decrease as the temporal and spatial resolution of the event increases and as the uncertainty decreases. Fig. 2 The general pathway of factors leading to a Rift Valley fever outbreak ElNiño/IOD Rainfall Vegetationgrowth+Floods Vectorpopulationincrease (primaryfocalareas) Vectordispersal IncreasedRVF circulationinanimals Fullextentofthe RVFepizootic Animalmovements Implicationofothervectors Dashed blue boxes indicate key factors to be monitored for RVF outbreak risk mapping. Solid red boxes indicate where existing models are connected. – 15 – Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 Fig. 3 A roadmap for Rift Valley fever forecasting – 16 – Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 VII. Recommendations FAO representatives, WHO representatives and the participants of the Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models put forth the following recommendations for the development of RVF forecasting and real-time outbreak management tools: 1. The accuracy of RVF potential major outbreak area maps should be increased in order to improve forecasting models. Statistical models rely on data availability, data accuracy and their ecological significance. To improve the quality of RVF potential major outbreak area maps, FAO and WHO should coordinate the development of a common and comprehensive database of geolocations of RVF animal and human cases that will be made available to international partners and scientists. This database should be checked for accuracy by field investigators, including FAO and WHO. It should include GPS data of primary emergence sites during major RVF outbreaks and of secondary epidemic sites, for example, irrigation systems where the disease is later introduced. 2. Forecasting models should be further developed to be more specific. To decrease the number of false alarms and be in a position to target alert messages to the prediction of only major RVF outbreaks, the specificity of RVF forecasting models should be increased as follows: 2.1. The RVF potential major outbreak area maps should be improved with soil type data, elevation data, vector distribution maps, domestic ruminant distribution maps and geolocations of major RVF outbreaks (see recommendation 1). 2.2. Taking into account the successful prediction of the RVF outbreak in the Horn of Africa and the Sudan in 2006, the NASA system based on NDVI data could be improved and adjusted in the following ways: x Rainfall should be explicitly incorporated into the algorithm for risk assessment. The input of rainfall will provide a back-up check on the NDVI anomalies and allow for the improvement of RVF outbreak risk mapping by ranking risk based on accumulated rainfall. There are many different rainfall estimate products derived from satellite images. Each has specific accuracy according to the region where they are used. Different daily rainfall estimate products should be evaluated so the most appropriate one for the region of interest can be used. Examples of rainfall estimate products include: i) the RFE rainfall product from the NOAA/Climate Prediction Center, ii) NOAA/Climate Prediction Center CMORPH rainfall products and iii) NASA TRMM rainfall estimates. x The RVF Potential Epizootic Area Mask (PEAM) used by the NASA system needs to be improved. Some of the RVF outbreaks along coastal Kenya in 2006/07 and recently in South Africa (January – February 2008) were outside the PEAM. An improvement in the PEAM will involve either a change in the rainfall and NDVI threshold values or incorporation of a more detailed land cover map and will require collaboration with insights from FAO, WHO and in-country experts. The outcomes of the model developed by Oxford University’s TALA Research Group should also be considered. x Results from RVF sentinel sites could be used to improve the RVF potential major outbreak area maps and the RVF PEAM. A number of sites which are foci or epicentres of recent and previous RVF outbreaks should be identified and used to monitor rainfall and NDVI, to serve as area specific indicators of early warning information. The plotted – 17 – Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 series from the RVF sentinel sites could be published along with NASA’s monthly updated maps to provide additional value added information to the response planning component. In the sentinel sites RVF virus activity should be concurrently monitored in animal and mosquito populations. 2.3. The database described in recommendation 1 should also be used to run the model on RVF suitability mapping developed by Oxford University’s TALA Research Group, with a clear distinction between primary and secondary sites. 2.4. Forecasting model experts together with FAO, WHO and end-users should develop guidelines to help with interpretation of model outputs and decision taking regarding when to decide that an outbreak is imminent and when to alert countries. 3. Models should be coupled to buy time and precision; an alert signal should be sent six months before the start of an animal outbreak. 3.1. Considering the roadmap for RVF forecasting discussed during this meeting (Fig. 3), the use of existing climate forecasting models including IOD and SST should be further explored with the final objective of linking them with NDVI based forecast models and RVF epidemic risk models. 3.2. Flood maps should be included in the risk assessment. A USGS-like approach would be highly valuable and help to identify vulnerable areas and real-time satellite data to determine flooded areas. A pilot study on the map of a flooded area, Kenya 2006–2007, could be done to explore the capacities for real-time monitoring of flooding and to investigate the link between flooding and RVF emergence. This study would benefit from the highly consistent work already conducted during the USGS FEWS NET and IRI project. 3.3. In such flooded areas, epidemiological activities, including the detection of RFV virus activity in animals, the surveillance of mosquito populations and the risk of dissemination from these areas through animal movement or the passive spread of mosquitoes, must be supported. 4. RVF forecasting models and potential major outbreak area maps should be used to monitor the evolution of RVF outbreaks over time. The prediction of the end of an outbreak based on climatic factors is an important tool for ministries of agriculture and ministries of health, to enable them to prepare for the end of the outbreak and scale down field operations. 4.1 Ministries of agriculture and veterinary services should monitor and report movements of domestic ruminants during RVF outbreaks. At this stage, forecasting models only use static animal data. There is a significant amount of animal movement during an RVF outbreak, which directly affects the spread of the disease, but the data available on this are very broad. 4.2 The possible contribution of wind to the spread of infected mosquitoes during RVF epidemics should be evaluated. The database developed in recommendation 1 should be used to analyse trajectories in light of the chronology of events to verify if wind plays a role in the spread of RVF during major outbreaks. – 18 – Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 5. The participation of ministries of meteorology, ministries of health, and ministries of agriculture in the forecasting alert system should be enhanced. 5.1. FAO and WHO should inform their Member States through their RVF web sites that monthly updated RVF outbreak risk maps are posted on the US Global Emerging Infections Surveillance and Response System (DoD-GEIS) web site 10 . 5.2. FAO, WHO and their partners should organize regional training sessions and workshops on interpreting RVF risk updates on a monthly basis and analysing/assessing forecast alert messages. 5.3. Current efforts should focus on the most affected countries: 10 x Primary: Ethiopia, Kenya, Madagascar, Somalia, South Africa, the Sudan and the United Republic of Tanzania. x Secondary: the Comoros, Egypt, Mauritania, Mayotte (France), Saudi Arabia, Senegal, Swaziland and Yemen. http://www.geis.fhp.osd.mil/GEIS/SurveillanceActivities/RVFWeb/indexRVF.asp – 19 – Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 Acronyms AU-IBAR AVHRR BERM CDC DEM DoD-GEIS ENSO EVI FAO GADS IOD IRI ITCZ LST LSWI MODIS NASA NDVI NOAA OIE OLR PEAM RFE RSLTC RVF SSM SST UNICEF USAID USGS FEWS NET WHO WIO African Union/Inter-African Bureau for Animal Resources Advanced very high resolution radiometer Basin excess rainfall maps Centers for Disease Control and Prevention, Atlanta, USA Digital elevation model Department of Defence-Global Energizing Infections Surveillance and Response System El Niño/Southern Oscillation Enhanced vegetation index Food and Agriculture Organization of the United Nations Global Data Assimilation System Indian Ocean Dipole International Research Institute for Climate and Society, USA Intertropical convergence zone Land surface temperature Land surface water index Moderate resolution imaging spectroradiometer National Aeronautics and Space Administration Normalized difference vegetation index National Oceanic and Atmospheric Administration World Organisation for Animal Health Outgoing long wave radiation Potential epizootic area mask Rainfall estimates Red Sea Livestock Trade Commission Rift Valley fever Sentinel Site Monitoring Sea surface temperatures United Nations Children’s Fund United States Agency for International Development United States Geological Survey Famine Early Warning Network World Health Organization Western Indian Ocean – 20 – Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 Participants, observers and FAO-WHO Secretariat Participants Dr Assaf Anyamba NASA Goddard Earth Sciences Technology Center, University of Maryland Baltimore County, the United States of America Dr Pietro Ceccato International Research Institute for Climate and Society, The Earth Institute, Columbia University, New York, the United States of America Dr Glyn Davies Whittery, Chirbury, Montgomery powys, the United Kingdom Dr Lorenzo De Simone EMPRES/Animal Production & Health Division (AGAH), FAO, Rome, Italy Dr Hussein Gadain FAO Somalia Water and Land Information Management - SWALIM Project, Nairobi, Kenya Dr Marius Gilbert Lutte Biologique et Ecologie Spatiale (LUBIES), Université Libre de Bruxelles, Brussels, Belgium Dr Guy Hendrickx Avia-GIS, Zoersel, Belgium Dr Kennet J. Linthicum Center Director, USDA-Center for Medical, Agricultural & Veterinary Entomology, Gainesville, Florida, the United States of America Dr Vincent Martin FAO representation in China, Beijing, China Dr Tim Robinson Animal Production & Health Division (AGAL), FAO, Rome, Italy Dr David J. Rogers Oxford University, TALA Research Group, Dept. of Zoology, Oxford, the United Kingdom of Great Britain and Northern Ireland Dr Jan Slingenberg Animal Production & Health Division, FAO, Rome, Italy Dr Chun Knee Tan United Nations University, Global Environment Information Centre (GEIC), Tokyo, Japan Dr Xiangming Xiao Complex Systems Research Center, Institute for the Study of Earth, University of New Hamsphire, Durham, the United States of America – 21 – Joint FAO-WHO experts consultation on Rift Valley fever outbreak forecasting models FAO headquarters, Rome, 29 September–1 October 2008 Observers Dr Yehia Ghazi OIE Regional Animal Health Center, Beirut, Lebanon Dr Castiello Massimo FAO Special Emergency Programmes Services, Mogadishu, Somalia FAO-WHO Secretariat Dr Stéphane de La Rocque EMPRES / Animal Production & Health Division (AGAH), FAO, Rome, Italy Dr Pierre Formenty Global Alert and Response Department (HSE/GAR), WHO, Geneva, Switzerland – 22 –